Abstract: Existing multi-label learning algorithms mostly assume that all class labels in an obtained data set are completely accurate. This assumption is rarely valid. As the scale of the label space ...